Will we be able to better communicate with bots in the future? Google is working on an AI language model that makes bots understand context.
Interaction between humans and robots has become the most natural thing today. We talk to Google Assistant and Siri about things in life, and more and more companies are using chatbots to help their customers. Those who remember the early days of chatbots will agree that communication with bots has become more natural, although limitations in the language understanding of bots remain. Google is trying to fix its biggest flaw: understanding context.
in a blog Google shares the latest developments in the research project PaLM-SayCan, an advanced artificial intelligence language model developed for assistive robots. This language model should not only enable bots to read between the lines and no longer interpret our questions one hundred percent literally, and then associate the correct action(s) with these on an independent basis. Google is already fully testing this language model in practice on Everyday Robots.
Google fires AI engineer who claims chatbots are conscious
Snack and drink
With some daily examples, Google shows how PaLM-SayCan can provide a new breakthrough in human-robot interaction. Let’s say you tell a bot that you want a snack and “something prompts it to throw it away”. Especially with the second part of the question, today’s bots will have a hard time using images. However, PaLM-SayCan can make sure the bot understands that you need a matching snack and drink.
exceed that. If you pour the drink and ask the robot to clean it, it should be able to decide for itself to buy a sponge and not a vacuum cleaner. The combination of getting a drink first and then putting the empty can in the trash shouldn’t be a problem.
The first trials with PaLM-SayCan have already shown promising results. Google writes that the language model increased the planning ability of bots by 14 percent and the success rate of actions by 13 percent. The most promising result is that robots have a 26 percent better ability to complete more complex tasks consisting of multiple actions. Of course, many experiments are necessary before the language model can be implemented on a larger scale. In addition, Google promises to respect the ethical guidelines for the development of artificial intelligence. For now, it appears that the use of PaLM-SayCan will remain limited to everyday applications.
“Lifelong entrepreneur. Total writer. Internet ninja. Analyst. Friendly music enthusiast.”